Skip to content

device-lynx 0.2.2: accept devinfo 3.11 for boot-success - #436

Merged
derekross merged 9 commits into
derekross:mainfrom
cx3301J:lynx-bootctl-3.11
Nov 13, 2026
Merged

derekross merged 9 commits into
derekross:mainfrom
cx3301J:lynx-bootctl-3.11

Conversation

@cx3301J

@cx3301J cx3301J commented Nov 5, 2026 •

Copy link
Copy Markdown
Contributor

Android 15 lynx units report devinfo 3.11, while omamobile-device-lynx 0.2.0 only accepted 3.15. On this unit omamobile-boot-success refused to write, so slot B stayed unmarked and burned its retries.

The A/B slot bytes are at the same offsets in both layouts. This adds a verified-layout table for 3.11 and 3.15, keeps the write refusal for anything else, and covers the 3.11 one-byte plan plus an unverified 3.14 rejection in the lynx package test.

  • Version is 0.2.2 because 0.2.1 is already on main.
  • device/lynx/NOTES.md records the unit evidence; the matrix row notes 3.11 support from 0.2.2.
  • On the phone: published kernel 7.3.0-rc6-omamobile-push1-gfbbcac19d8ea, GT9895 touch works, slot B's earlier marking is not recorded.

Package tests pass, including ok - bootctl accepts devinfo 3.11 and ok - bootctl refuses an unverified devinfo layout.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you for this, and for writing up the 3.11 unit so clearly. A phone whose bootloader is a release behind is exactly the case our boot-success refusal was built for, and I'm glad it refused rather than guessed. I'd like to get 3.11 supported, but not quite this way yet:

  1. The Pixel 7 Pro's copy changes too. Lynx's test keeps its omamobile-bootctl.c identical to cheetah's, so the PR edits both, and the 7 Pro would start writing devinfo on any 3.11 header on the strength of a lynx bootloader. The reference phone stays unchanged unless there's a reason that holds on it. Could you keep the shared source at 3.15 and have lynx's build add 3.11 at compile time (for example -DOMAMOBILE_DEVINFO_EXTRA_MINOR=11, with an #ifdef next to the table)? Then cheetah's and raven's binaries stay byte for byte the same, and lynx's test can still compare the source.
  2. The evidence for 3.11 needs the same table 3.15 had. For 3.15 we compared bytes 0x30-0x37 with fastboot getvar slot-successful, slot-unbootable and slot-retry-count for the same state, then again after a change the bootloader made itself (one unmarked boot); it's in device/lynx/BRINGUP-M1.md under "devinfo against fastboot". Could you add that table for the 3.11 unit? Please post only those eight bytes and the version, never a full dump.
  3. The write that already happened. The notes say slot B was marked with this build before it was reviewed. That's a devinfo write on a layout our tools didn't accept yet, so I'd like it recorded properly: which build wrote it, the eight bytes before and after, and whether you had a read-only devinfo backup first. From here on, please don't write devinfo on a layout that isn't in the merged table without checking with me first.
  4. Version: main already ships device-lynx 0.2.2 (it came with the release-channel change), and device-lynx 0.2.3: persist the DSIM transmit enable #438 takes 0.2.3, so this one gets the next free number when it merges. I'll set it and the title then.
  5. DEVICES.md: the row also turns Bluetooth from "WIP ... scans through its own UART" into "no", which I don't think you meant. The kernel side (linux-omarchy-mobile#98) merged on the 3rd, so that cell should stay as it is. Please keep only the System and touch changes, and make the System cell name the version that ships.

Smaller things: the comment above the new table still says 3.15 is the only layout; the new comment says 3.15 came from "the Android 17 bootloader used by the published lynx notes", which the notes don't say (3.15 was first checked on the 7 Pro); a tools/phone-smoke report from the unit would help; and the commit body wants wrapping at 72 columns. Please also note in NOTES that both slots carry lynx-15.2.

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

Thanks for the careful review, and for the 3.11 write concern. Agreed on all points.

  • I will keep omamobile-bootctl.c shared at 3.15 and add lynx's 3.11 only at compile time (define + #ifdef), so cheetah/raven binaries stay byte-identical.
  • I will add the 3.11 eight-byte table against slot-successful/slot-unbootable/slot-retry-count, including one bootloader-made change, and record the earlier write (build, before/after bytes, backup state).
  • No further devinfo writes on an unmerged layout; I will check with you first.
  • I will fix the 3.15 comments, keep Bluetooth as WIP in DEVICES.md, restrict the row to System/touch, and add the phone-smoke report and 72-column commit bodies.

I will bump the version only when you say which number is free. The phone is currently back on stock Android, so the table and smoke report need one Arch reflash after the patch is review-shaped.

Assisted-by: DeepSeek

@cx3301J

cx3301J commented Nov 6, 2026

Copy link
Copy Markdown
Contributor Author

Revised as requested:

  • The shared omamobile-bootctl.c table is back at 3.15 only; lynx adds 3.11 through -DOMAMOBILE_DEVIINFO_EXTRA_MINOR=11 and an #ifdef next to the table, so cheetah/raven keep the same compiled layout.
  • Rebased onto current main (cd90d19); the DEVICES conflict is resolved by keeping main's newer rows and changing only lynx's System/touch cells. Bluetooth stays WIP.
  • Version is 0.2.4; if 3.11 lands after another bump I will follow your number.
  • Commit has Signed-off-by and 72-column body.

Still to add from your list: the 3.11 eight-byte table, the record of the earlier write, the comment fixes, ools/phone-smoke, and the NOTES line that both slots carry lynx-15.2. Those need the one Arch reflash; I will do them before asking for another review.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you for turning this around so quickly, and for keeping the shared table behind a build-time switch as I asked. It's the right shape. I also appreciate that you didn't write devinfo again while 3.11 wasn't in the merged table. Two things still stand in the way, and one is on me for how I described the switch:

  1. The Pixel 7 Pro's binary still changes. I built cheetah's omamobile-bootctl.c from main and from your head with the same flags. The logic is the same, but the binary differs, and on a 7 Pro the refusal message would now say "not a verified layout (3.11, 3.15)", which isn't true there. A smaller change keeps it byte for byte the same (I checked): leave the VERIFIED_VER_MAJOR/VERIFIED_VER_MINOR lines and the old message as they are, and only extend the existing condition:

    if (for_write && (maj != VERIFIED_VER_MAJOR || min != VERIFIED_VER_MINOR)
    #ifdef OMAMOBILE_DEVINFO_EXTRA_MINOR
        && (maj != 3 || min != OMAMOBILE_DEVINFO_EXTRA_MINOR)
    #endif
        ) {

    (The macro in the PR is spelled DEVIINFO; please use DEVINFO.) I'll restage the 7 Pro's boot files after the merge, since any edit under its src/ needs that.

  2. The 3.11 bytes don't match the 3.15 layout yet. Read with the 3.15 layout, 03 00 00 00 00 0c 00 00 gives slot b a retry count of 0, but the getvar lines next to it say slot-retry-count:b 1. On 3.15 the retry byte matched getvar in every row of the table in device/lynx/BRINGUP-M1.md, and one unmarked boot visibly spent a retry. Here the bytes stayed the same across that boot. Slot a also reads 00 flags where 3.15 had 0a. Maybe the reads were taken at different moments (fastboot before the boot, Linux after it), or maybe 3.11 really is laid out differently. Either way, could you redo it in the 3.15 table's form: three rows (B marked, --set-active=b before booting, after one unmarked boot), each with the eight bytes and getvar slot-successful, slot-unbootable and slot-retry-count, and when each read was taken? Please still post only those eight bytes and the version.

  3. The earlier write. The 10-30 section still says a write changed 0x35 from 04 to 06, and the new section says no write was forced. Could you reconcile the two? Please say which build made that 10-30 write, give the bytes before and after, and say whether you had a read-only devinfo backup first.

  4. A safety note. If slot b still reads retry 0 and unmarked, the next reboot may fall back to stock on slot A. If recovery offers a factory reset, decline it (Volume Down + Power gets you to fastboot). You mentioned the phone was on stock Android on the 6th. Was that a fallback, or did you choose it? Either way, please take a fresh read-only backup (tools/nv-backup) before any radio work, because stock rewrites efs, persist and modem_userdata.

Smaller things: NOTES is newest first, so the 11-06 section belongs at the top, and the 10-30 "Fix:" bullet still describes the shared table. The title still says 0.2.2. I'll set the version and title at merge, since #98 and #438 also bump device-lynx. #98 is closest to ready, and #438 waits for the kernel side.

Thanks again. The package tests pass here, the 3.14 refusal test is a good addition, and tools/pr-check is clean.

@cx3301J

cx3301J commented Nov 9, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto current main (71b47a6) and followed your smaller shape exactly:

  • The shared source is back to the VERIFIED_VER_MAJOR/VERIFIED_VER_MINOR lines and the old refusal message; the only change is the #ifdef clause inside the existing or_write condition.
  • Macro spelling is OMAMOBILE_DEVINFO_EXTRA_MINOR, including the compiler flag.
  • Built cheetah's source from main and from this head with the same filename and flags; the binaries are byte-identical (SHA-256 d32f6448.).
  • Package tests pass, including 3.11 acceptance and 3.14 refusal.

On the bytes: you were right that the table was not same-state. I marked it provisional in NOTES.md - the Linux bytes and bootloader getvar were read in separate cycles, the retry byte disagrees, and slot A's flags differ. I have not claimed 3.11 is validated; a same-moment table still needs one Arch run.

#436 is mergeable/clean now.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you for the rework. The shared source now has exactly the shape we discussed, and I reproduced your check: cheetah's omamobile-bootctl from main and from your head build to the same binary (d32f6448…). The 3.14 refusal test is a good addition, and the package tests pass here. Thank you also for marking the 3.11 table as provisional instead of presenting it as more than it is.

That honesty is also why I can't merge it yet:

  1. The build still turns 3.11 writes on. Lynx's build passes -DOMAMOBILE_DEVINFO_EXTRA_MINOR=11, so with this merged, every Pixel 7a on a 3.11 bootloader would write byte 0x35 of devinfo at every boot, on a layout your notes say isn't confirmed yet. I'd rather not ship that write until the table is settled. Two ways forward, whichever you prefer: keep this open until the table is done, or take the -D…=11 out of build for now, so this PR lands the neutral #ifdef and the refusal test, and turn 3.11 on in a follow-up with the table. The 3.11 dry-run test can build its own copy with the define in $tmp. In that case the DEVICES cell would wait for the follow-up too.

    On the table itself: I think the 0-versus-1 retry mismatch may just be timing. The bootloader spends a retry each time it boots an unmarked slot, so a getvar taken before a boot and bytes read in Linux after it will be one apart even if the layout is right. For each row, read the eight bytes in Linux, then go to fastboot with the buttons (Volume Down + Power for about 20 s, since the bootloader often ignores reboot-to bootloader), and take getvar slot-successful, slot-unbootable and slot-retry-count before anything boots again. Rows for "after --set-active=b and one boot" and "after a second unmarked boot" need no write from Linux. The "B marked" row does, and that one is my call, as your notes say.

  2. The earlier write is gone from the notes rather than explained. The previous versions said, under 10-30, that marking slot b successful changed 0x35 from 04 to 06 and the bootloader kept the slot. The PR description still says "slot B marked successful with this build". The new text says no write was forced. Please put the 10-30 write back, with which build of omamobile-bootctl made it, the eight bytes before and after, and whether you had a read-only devinfo backup first. If you didn't have one, that's fine to say. And if the earlier text was wrong and nothing was written, say that instead and fix the description.

  3. The stock boot on the 6th. Was that a fallback or your choice, and from which slot? If stock Android has booted since your last backup, please take a fresh read-only one before any radio work, because stock rewrites efs, persist and modem_userdata.

Smaller: NOTES is newest first, so the 11-06 section belongs at the top. If you like, copy the same source change into the Pixel 6 Pro package's src/omamobile-bootctl.c too, so all three copies stay equal (its binary doesn't change either). I'll set the version and title at merge, and refresh the 7 Pro's staged files afterwards, since its source file changes. Thanks again for sticking with this. A bootloader that's one release behind is exactly the case worth getting right.

@cx3301J

cx3301J commented Nov 10, 2026

Copy link
Copy Markdown
Contributor Author

Took your second option: the shipped �uild no longer passes -DOMAMOBILE_DEVINFO_EXTRA_MINOR=11. The shared #ifdef and the 3.14 refusal stay; the 3.11 dry-run compiles its own copy in $tmp with the define, so the package test still covers it. The DEVICES System cell is back to the existing wording until the follow-up.

I also restored the 10-30 write record: omamobile-bootctl from omamobile-device-lynx 0.2.0 changed �x35 from �x04 to �x06; only that byte was recorded, not the full eight, and no read-only devinfo backup was taken first. The 11-06 section stays explicitly provisional with the retry mismatch.

On the same-moment table: I will capture the Linux bytes and then the getvar lines with the buttons before any further boot, including the two unmarked-boot rows, and leave the marked row for your call.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, this is the shape I asked for. Lynx's build no longer passes the define, the 3.11 dry run builds its own copy, DEVICES.md is left alone, and I reproduced again that the Pixel 7 Pro's omamobile-bootctl builds to the same binary from main and from your head (d32f6448…). The shipped lynx build refuses a 3.11 header here, with and without --i-understand. So the code is ready to land as an inert switch. Three small things first:

  1. One test line fails a check that landed on main today. bin/omamobile, tools, rootfs: no pipe into grep -q under pipefail #587 added tests/grep-q-pipes: under pipefail, a pipe into grep -q can fail at random, so it's no longer allowed in tests. Line 100 of test/all ("$tmp/bootctl311" dump … | grep -q 'version: 3.11') trips it once your branch is merged with main. Please rebase and make it | grep -c 'version: 3.11' >/dev/null, like the 3.15 line above it now is. (CI didn't show you this: your last push is waiting for my approval to run, which I'll give after the rebase.)

  2. The 10-30 write record contradicts the bullet above it. The notes say 0.2.0 only accepted 3.15, "so omamobile-boot-success refused to write and the slot was left unmarked", and two bullets later that omamobile-bootctl from 0.2.0 changed 0x35 from 04 to 06. Both can't be true: 0.2.0's tool has the same 3.15-only check and refuses a 3.11 header. Also, your unit's own bytes on 11-06 have 0c at 0x35, while 04 -> 06 is the example from our test and from the 3.15 table. My guess is that the write came from the first build of this PR (the description still says "slot B marked successful with this build"), but I'd rather not guess. Could you say which binary made the write, and whether 04 -> 06 is what the tool printed on your phone or was filled in afterwards? If you no longer know, "not recorded" is a fine answer; it just has to be what happened. Please update the PR description to match.

  3. The 3.14 test can't fail. A dry run of mark-successful always exits 1, accepted or not, so if … mark-successful …; then fail never fires. I checked by building a copy that accepts 3.14: the test still passes. Could you capture the output instead and check for refusing to write (and that bytes to change is absent)? And please add the same check for the shipped build ($tmp/build/omamobile-bootctl) on your 3.11 image, plus a line that build doesn't mention OMAMOBILE_DEVINFO_EXTRA_MINOR. That is the property this PR now promises, so it deserves a test.

Smaller, and still open from before:

  • NOTES is newest first, so the 11-06 section belongs at the top, above 10-30.
  • The stock Android boot on the 6th: was it a fallback or your choice, and from which slot? If stock has booted since your last backup, please take a fresh read-only one (tools/nv-backup) before any radio work.

With those in, I'll merge this as device-lynx 0.2.3 (or the next free number that day) and fix the title. The follow-up that turns 3.11 on waits for the same-moment table you described (Linux bytes, then fastboot by the buttons, then getvar, before anything boots again), and the "B marked" row stays my call, with a read-only devinfo backup taken first this time. Thanks for sticking with it.

@cx3301J

cx3301J commented Nov 11, 2026

Copy link
Copy Markdown
Contributor Author

All three are in, head e629841 after a rebase onto c3cfd86:

  1. The 3.11 dump line uses | grep -c 'version: 3.11' >/dev/null; ests/grep-q-pipes passes on the rebased branch (172 scripts).
  2. The 3.14 test now captures the output and requires
    efusing to write with no �ytes to change. The shipped $tmp/build/omamobile-bootctl is checked against the 3.11 image (refuses with --i-understand), and the test fails if �uild mentions OMAMOBILE_DEVINFO_EXTRA_MINOR.
  3. I rewrote the notes rather than guessing: the 10-30 write is recorded as not recorded. The �x04 -> 0x06 value came from the 3.15 test/table, not a phone reading; 0.2.0 refuses 3.11, the binary that wrote slot B is unknown, and no read-only devinfo backup was taken first. The PR description no longer says this build marked slot B.

Also moved the 11-06 section to the top, and noted that the stock boot on the 6th was a deliberate restore from slot B (factory images), with no radio work after it.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. All three points from last time are in, and I checked each on the new head:

  • The 3.11 dump line no longer pipes into grep -q, and tests/grep-q-pipes passes on your branch.
  • The 3.14 test can fail now. I put the define back into build and the suite stopped at "the shipped build accepted 3.11"; I made the source accept 3.14 and it stopped at "an unverified 3.14 layout was not refused".
  • The notes say "not recorded" for the 10-30 write. That is the honest answer, and it is what the record needed.

The Pixel 7 Pro's omamobile-bootctl still builds to the same binary from main and from your head (d32f6448…), and the lynx and cheetah package tests pass here.

One thing in the notes before I merge. The new paragraph on the stock restore ends with "No radio work followed, so no fresh persist/efs backup was needed." Our rule is the other way round: after any stock Android boot a fresh read-only backup is taken, because stock rewrites efs, persist and modem_userdata, and without it the next compare can't tell stock's change from ours. Not having done radio work since only means it hasn't mattered yet. Could you change that sentence to say a fresh read-only backup (tools/nv-backup) is still owed before any radio, PMIC, AoC or Trusty work on this unit, or give the date if you have taken one?

And since the paragraph now says the factory images were flashed to slot B, please add three facts to it, because they matter for the 3.11 table in the follow-up:

  • what was flashed and how (Google's flash-all, which also writes the bootloader and radio and wipes userdata, or only the images to slot B), and which build;
  • what slot A holds now;
  • whether the 11-06 bytes and getvar lines above it were read before or after that flash, since a factory flash resets slot B's flags and retry count.

Two wording nits while you're there: the 10-30 section says "see the provisional table below", and the table is above it now; and "Both slots on this unit carry bootloader lynx-15.2" could say how the other slot's version was read, since this bootloader reports only the current slot's.

On the version: the package's installed files don't change with this PR (same binary, and the test isn't shipped), so I'll merge it without a version bump and write the title to match what it now does. You don't need to change the version file; I'll handle it at merge.

The follow-up that turns 3.11 on still waits for the same-moment table, as we agreed, and the "B marked" row stays my call with a read-only devinfo backup first. Thanks for staying with this through five rounds.

@cx3301J

cx3301J commented Nov 12, 2026

Copy link
Copy Markdown
Contributor Author

Fixed in �ab3112: the note no longer says a backup was unnecessary. It now says stock rewrites efs, persist and modem_userdata, so a fresh read-only backup is required before any radio work. I have not taken that backup yet; when the phone is available with adb I will run ools/nv-backup before touching any radio.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J. The sentence that said no backup was needed is gone, and that was the one I minded most. I checked the new head: the code and the test are the files I went through last time (only NOTES changed), the lynx package test passes on a merge with today's main, the shipped build still refuses a 3.11 header, and the Pixel 7 Pro's omamobile-bootctl builds to the same binary from main and from your branch, both as the package builds it and as the boot image does.

One thing is needed before I can merge, the first below. The other two can come in a follow-up pull request of their own, since the code here stays off either way, but please do send them:

  1. The branch conflicts with main now, in packages/omamobile-device-lynx/version (main moved to 0.2.3 while yours says 0.2.4). As I said last time, this PR needs no version, so please take main's file: git fetch origin && git merge origin/main, then git checkout origin/main -- packages/omamobile-device-lynx/version and commit. GitHub doesn't run CI on a conflicting PR, so that also gets the checks going.

  2. The three facts about the factory flash on the 6th, in the same NOTES paragraph. They went unanswered last time, and they matter more than they look:

    • what you flashed and how (Google's flash-all, which also writes the bootloader and the radio and wipes userdata, or only the images to slot B), and which build;
    • what slot A holds now;
    • whether the 11-06 bytes and getvar lines were read before or after that flash.

    If it was a full flash-all of a newer build, the bootloader was replaced, and the phone may no longer report devinfo 3.11 at all, which would change what the follow-up has to do. If you don't remember, "not recorded" is a fine answer, as it was for the 10-30 write.

  3. The backup sentence, a little further. Please say that the fresh read-only backup has not been taken yet, and that it comes before any radio, PMIC, AoC or Trusty work on this unit. One correction to your reply: tools/nv-backup runs from our Linux over USB SSH, not through adb. Android gives you no root to read those partitions, and booting stock again would change them once more.

Two small wording points from before, while you're in the file: the 10-30 section says "see the provisional table below" and the table is above it now, and "Both slots on this unit carry bootloader lynx-15.2" could say how the other slot's version was read, since this bootloader reports only the current slot's.

The title and description still describe the first version (accepting 3.11, a matrix note); I'll write the merge title to match what it now does, and it would help if the description said the same. The follow-up that turns 3.11 on still waits for the same-moment table, and the "B marked" row stays my call, with a read-only devinfo backup first.

Android 15 lynx units report devinfo 3.11, so the boot-success marker
refused to write. Keep the shared verified-layout table at 3.15 and add
3.11 only for lynx through a compile-time define, leaving the 7 Pro and
raven binaries unchanged.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
Add the measured eight bytes at 0x30-0x37 and the matching fastboot
slot state for the unmarked Android 15 unit, plus the note that both
slots carry bootloader lynx-15.2. devinfo is read-only on this unit, so
no write was forced while 3.11 is not in the merged table.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
The condition uses OMAMOBILE_DEVINFO_EXTRA_MINOR; pass the same spelling
to the compiler so lynx actually accepts 3.11.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
The eight bytes and the fastboot lines were read at different moments, and
the retry byte disagrees with getvar. Keep the observation but do not call
it a same-state table; a same-moment capture is still needed.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
Remove the build-time define from the shipped package, keep the shared
ifdef, and let the 3.11 dry-run compile its own copy. Restore the 10-30
write record with the build and the one byte that was captured.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
Rebase onto today's main, make the 3.11 dump check pipefail-safe, and
test that 3.14 is refused and the shipped build refuses 3.11. Correct the
10-30 write record to say it was not recorded, move the newest notes to
the top, and note that the stock boot on the 6th was a deliberate restore.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
Stock rewrites efs, persist and modem_userdata, so the note now says a
fresh read-only backup is required before radio work instead of saying
one was not needed.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Contributor Author

Rebased onto current main (12258d9). Main has since moved to �.2.4 itself, and the branch's version file is now identical to main's, so the conflict is gone without a version change.

Head 59137cf. The two follow-ups (copy the source to the Pixel 6 Pro package, same-moment table) remain separate, as you suggested.

Assisted-by: DeepSeek

The Linux side of a same-moment row: slot B after a boot read the eight
bytes as 03 00 00 00 01 0c 00 00, with the retry byte matching getvar's
1. The fastboot side still has to be taken before the next boot.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Contributor Author

Recorded the Linux side of a same-moment row in ca75e21:

slot B, after a boot, devinfo 0x30-0x37: 03 00 00 00 01 0c 00 00 blockdev --getro: 1

Retry byte is 1, matching the getvar that said 1 last time. The fastboot side still has to be taken before the next boot to complete the row.

Assisted-by: DeepSeek

The bytes and the matching getvar were read without a boot between
them: retry 1 in both, slot B unmarked and bootable. This is the row the
3.11 table needs; the write stays off in the shipped build.

Signed-off-by: cx3301J <cx3301J@users.noreply.github.com>
Assisted-by: DeepSeek
@cx3301J

cx3301J commented Nov 13, 2026

Copy link
Copy Markdown
Contributor Author

Completed the same-moment row without a boot between the reads, in �473c43:

`
Linux, slot B, devinfo 0x30-0x37:
03 00 00 00 01 0c 00 00

bootloader, before anything booted:
slot-successful:b no
slot-unbootable:b no
slot-retry-count:b 1
current-slot: b
`

Retry �x34 is 1 in both; �x35 is �x0c. The shipped build still does not pass the 3.11 define, so this changes no write.

Assisted-by: DeepSeek

@derekross

Copy link
Copy Markdown
Owner

Thank you, @cx3301J, and thank you for the same-moment row. That is exactly the read I hoped for: the Linux bytes first, then fastboot by the buttons, then getvar, with nothing booting in between. Read with the 3.15 layout, slot b's 01 0c is retry 1, active, not successful, not unbootable, and all four match the getvar lines you posted. So the earlier 0-versus-1 mismatch looks like timing, not a different layout. That is real progress for the follow-up.

For this pull request, I checked the new head. The code and the test are the files I reviewed before (only NOTES changed), the lynx package test passes, the shipped build still refuses a 3.11 header, and the Pixel 7 Pro's omamobile-bootctl still builds to the same binary from main and from your branch. Since main moved device-lynx to 0.2.4 by itself, the version conflict is gone, and as said before, this PR needs no version of its own. Once CI has run (I'll approve it), I'll merge it with a title that says what it now does.

For the follow-up that turns 3.11 on, here is what's still open. None of it holds up this PR:

  1. The factory flash on the 6th, in the NOTES paragraph: what was flashed and how (Google's flash-all, which also writes the bootloader and radio and wipes userdata, or only the images to slot B), and which build; what slot A holds now; and whether the 11-06 reads came before or after that flash. "Not recorded" is a fine answer for any of them.
  2. Slot B has one try left. Your row shows slot b unmarked with retry 1. The next unmarked boot of B uses that try, and the boot after it falls back to slot A, which your bytes show unmarked with 3 tries. So the "after a second unmarked boot" row is the last one B can give before a fallback. Please answer point 1 before taking it, so we both know what slot A would boot. If stock comes up and recovery offers a factory reset, decline it (Volume Down + Power held about 20 s gets you to fastboot).
  3. Slot a's lines too. In the next rows, please add getvar slot-successful:a, slot-unbootable:a and slot-retry-count:a next to the bytes, so slot a's 03 00 gets compared as well.
  4. The backup sentence. Please say that the fresh read-only backup hasn't been taken yet, and that it comes before any radio, PMIC, AoC or Trusty work on this unit.
  5. Two wording points. The 10-30 section says "see the provisional table below", but the table is now above it. And "Both slots on this unit carry bootloader lynx-15.2" could say how the other slot's version was read. It would also help to add one line under the old provisional paragraph saying that the 11-13 row, read without a boot in between, agrees on the retry count.

The "B marked" row stays my call, with a read-only devinfo backup taken first. Thanks again for your patience through all of this.

@derekross
derekross merged commit 8f84243 into derekross:main Nov 13, 2026
2 checks passed
derekross added a commit that referenced this pull request Nov 13, 2026
…#738)

The release kernel kernel-2026.11.13 (7.3.0-rc6-omamobile-g36d61a33795b)
booted on the reference phone from images built from this tree
(device/cheetah/NOTES.md, 2026-11-13), and #436 changed
src/omamobile-bootctl.c, which the staged boot files hash. A new version
for the new boot files; no source change.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants